Matrix: 2D array operations
- Create
- Get
- Update
- Map
- Reorder
- Aggregate (per column)
- Shrink
- Grow
- Reshape
- Convert
Create
| What | How | Details |
| Undefined (size n x m) | np.empty((n, m)) | Warning: don’t use the initial values |
| Zeros (size n x m) | np.zeros((n, m)) | |
| Ones (size n x m) | np.ones((n, m)) | |
| True values (size n x m) | np.full((n, m), fill_value=True) | |
| False values (size n x m) | np.full((n, m), fill_value=False) | |
| Fill with value v (size n x m) | np.full((n, m), fill_value=v) | |
| Identity matrix of size n x m | np.eye(n, m) | |
| Diagonal matrix from vector | np.diag(v) | |
| From vectors as rows | np.row_stack((v1, v2)) | |
| From vectors as rows | np.vstack((v1, v2)) | |
| From vectors as rows | np.array([v1, v2]) | |
| From vectors as columns | np.column_stack((v1, v2)) | |
| From vectors as columns | np.stack((v1, v2), axis=1) | |
| From vectors as columns | np.vstack((v1, v2)).transpose() | |
Get
| What | How | Details |
| Element at row i, column j | m[i,j] | |
Update
All operations are in-place.
| What | How | Details |
| Update element at row i, column j | m[i,j] = v | |
| Fill with scalar value | m.fill(v) | |
| Fill with array values | m[...] = np.array(3) | |
| Fill row i with value | m[i] = v | |
| Fill column j with value | m[:,j] = v | |
| Resize to shape, fill with zeros | m.resize(m, d) | |
Map
Operations are element-wise and preserve the shape of the matrix.
| What | How | Details |
| Increment all elements | m + v | |
| Increment elements by the respective element of another matrix | m + m2 | Must be equal shape |
| Find min between two matrices | np.fmin(m, m2) | |
| Find max between two matrices | np.fmax(m, m2) | |
| Clip (truncate) between [a, b] | np.clip(m, a_min=a, a_max=b) | a_min < a_max is not checked |
Reorder
Operations preserve the shape of the matrix.
| What | How | Details |
| Reverse elements | np.flip(m) | Flattened view in reverse order |
| Sort elements descending | np.sort(m) | NaNs are last |
| Reverse column order | np.fliplr(m) | |
| Reverse row order | np.flipud(m) | |
| Sort elements by column descending | np.sort(m, axis=0) | |
| Sort elements per row descending | np.sort(m, axis=1) | |
Aggregate (per column)
Summarize along an axis. Set axis=1 for per-row operation.
| What | How | Details |
| Min | np.min(m, axis=0) | Use nanmin() to ignore NaNs |
| Max | np.max(m, axis=0) | Use nanmax() to ignore NaNs |
| Max - min | np.ptp(m, axis=0) | |
| Sum | np.sum(m, axis=0) | Use nansum() to ignore NaNs |
| Mean | np.mean(m, axis=0) | Use nanmean() to ignore NaNs |
| Median | np.median(m, axis=0) | Use nanmedian() to ignore NaNs |
Shrink
| What | How | Details |
| Diagonal vector | np.diag(m) | |
Grow
| What | How | Details |
| Pad with value | np.pad(m, pad_width=1, constant_values=v) | |
| Pad with edge element | np.pad(m, pad_width=1, mode='edge') | |
Reshape
| What | How | Details |
| Transpose | m.T | |
| Transpose | m.transpose() | |
| Rotate | np.rot90(m) | |
| Reshape to dimensions | m.reshape(d) | |
Convert
| What | How | Details |
| 1D array (concat rows) | m.flatten() | |
| 1D array (concat columns) | m.T.flatten() | More intuitive than specifying mode |
| 1D array (concat columns) | m.flatten(mode='F') | |
| List of rows | m.tolist() | |